Clearly defined neural network architecture
P. Sh. Geidarov · Optical Memory and Neural Networks · 2015
This article is dedicated to description of clearly defined neural networks on the basis of metric methods of recognition in which realization quantity of layers, neurons and connections are strictly determined according to the task initial conditions: quantity of images, samples and features. Also it describes the algorithm of training sample selection, as well problems related to opportunities of neural network architecture multitask application and opportunities of neural networks configuration optimization with usage of the clusterization algorithm applying to tasks of recognition and usage of network sequential recognition algorithm.